There is known an image generation system for generating an image visible from a given view point within an object space which is a virtual three-dimensional space. This image generation system is highly popular as a system which is capable of experiencing a so-called virtual reality. If such an image generation system is used for a gun game, a player (or operator) can enjoy a three-dimensional game by using a gun-type controller (or shooting device) to shoot target objects such as enemy characters (or objects) which are displayed on a screen.
In such an image generation system, it becomes an important technical problem to generate a more realistic image for improving the virtual reality. It is thus desirable that the motion of enemy characters can more realistically be represented as well. The image generation systems of the prior art have represented the motion of enemy characters by selecting a previously provided motion data and playing (replaying) a motion based on the selected motion data.
However, such a technique of playing the motion based on the motion data had the following problems:
(1) Even if an enemy character is hit, it only performs the same behavior at all times. This makes the representation of motion monotonous.
(2) If the enemy character is further hit, the play of motion started by hitting of the first shot (bullet) will be cut off by the second shot. This makes the motion of the enemy character unnatural.
(3) In order to increase the variation of motion for the enemy character, it is required to increase the motion data correspondingly. However, the capacity of the memory for storing the motion data is finite. Thus, the increase of the motion variation is also limited.